Regulatory information transformation ruleset expansion to support automated building code compliance checking

Abstract

The traditional building code compliance checking process mainly relies on design reviewers to review design documents or models manually. The intensive manual effort needed makes this process time-consuming, costly, and error-prone. Automated compliance checking (ACC) could be a promising upgrade of the traditional manual code compliance checking. With a reduced workload on design reviewers, ACC is cheaper, faster, and immune to human errors. To support ACC, building code requirements need to be represented in a computer-processable format to enable automated reasoning, which will in turn allow an automated assessment of the building design's compliance status with building codes. A major limitation of many existing ACC systems/methods is their limited range of checkable building code requirements. To address that, the state of the art uses pattern matching-based rules to transform building code requirements into computable formats automatically, but the ruleset was developed and tested only on a few chapters of building code requirements. An efficient ruleset expansion method is needed to increase its range of checkable building code requirements at a low cost, to bring ACC systems closer to full deployment. In this paper, the authors proposed a new regulatory information transformation ruleset expansion method for expanding an existing ruleset. This method can expand the range of checkable code requirements of ACC systems without significant manual effort. The proposed ruleset expansion method takes an iterative approach to ensure the generality and validity of new pattern matching-based rules and the quality of information transformation results. The expanded ruleset was tested on generating logic clauses from Chapter 5 of the International Building Code 2015. Compared to the baseline ruleset, the expanded ruleset increased the predicate-level precision, recall, and F1-score of the logic clause generation by 10.44%, 25.72%, and 18.02%, to 95.17%, 96.60%, and 95.88%, respectively.